2 replies to this topic

[ctfazra]

Briefly, I have 2 groups: healthy control & patients (n=9 in each group). Blood collected from each of this group before and after certain treatment. Then I went on to compare RNA expression, cell counts, etc between these groups, as well as before-after comparison. Controls are sex-matched.

This is how I analysed the results:

qPCR

1) Gene expression, normalised to housekeeping gene, before & after treatment (in each individual group) ---> Wilcoxon paired t-test (because data not normally distributed)

2) qPCR gene expression results were transformed into log2 ratio (after treatment/steady state) --> for this I used Mann-Whitney U test to compare ratios control vs. patients.


Cell counts

3) cell counts comparing steady state (pre-treatment) in controls vs. patients
4)  "     "            "            post-treatment     "             "             "             "  

----> for these two cell counts analysis (3 & 4) I used normal unpaired t-test (therefore, comparing the means).

5) cell counts comparing before & after treatment in each group (e.g. patients before vs. patients after)

----> this one I used normal paired t-test. However I'm not sure if I should use Wilcoxon instead. Is it possible for one group is normally distributed, while the other is not? If this is true, which test should I use?

6) Is there any other statistical test I can derive from the data I have? I'm asking this because my controls are sex-matched to the patients but the matched test I did was only on the individual group (before-after), not comparing both groups..if you get what I mean...

------------------------------------------------------------------------------        

Can anyone here help me to validate the statistical test I used in the above situations?

I've been asking around and reading loads about t-test, but it seems the more I read the more I get confused . I'm afraid if I've been doing the wrong statistical analysis all this, it'd be a huge waste of time.

Thanks in advance!                                 

[bob1]

Your sample numbers are too low to be using parametric tests like student's t-test, you should probably always be using non-parametric tests such as the wilcoxon.

[ctfazra]

OK, the main reason would be small sample number, therefore I should opt for non-parametric test in all the situations above. Thanks!